Software
Model
1.Land use change models
The proposed research will combine two types of models: land use change models and models of the spread of invasive species. Modeling land use changes is a rapidly growing scientific field. In fact, two of the senior personnel on this proposal (Silander and Allen) also have independently developed explanatory land use models in a spatial Bayesian framework that are driven by socio-economic and landscape physical factors.
In this project, we plan to develop one particularly promising approach, Markov chain models, for land use change analysis. We will compare the new developed Markov chain models with those developed by two of the senior personnel on this proposal (Silander and Allen) based on the spatial Bayesian framework. A large body of basic research using one dimensional (1-D) Markov chain models exists in land use change studies. However, there are limitations in the existing 1-D Markov chain models. Existing 1-D Markov chain models only deal with first-order processes and they cannot accommodate higher-order effects, the influence of exogenous or endogenous variables. The probability of change from the current land use state to another state may depend not only on the current land use state, but also previous states. Yet, 1-D Markov chain models use only one-step transition probabilities (i.e., transition probability matrices) to grasp dynamical characteristics of land use such as the diversity, driving forces, or scale dependence of land use. The problem is that the non-Markovian property of the data cannot be reflected in transition probabilities derived from methods based on the first-order Markovian assumption. In addition, when transition matrices for different observation periods are compared, the observation intervals often differ because satellite images or photographs of the research site taken at constant time intervals may not be available. If the observation intervals differ, the transition probabilities cannot be compared without calculating a transition matrix with the normalized observation interval. However, difficulties may arise when applying such calculation to a practical dataset. In a heterogeneous landscape, where the spatial relationships of natural and human dominated land uses are interspersed in different ways, the lack of spatial dependence in transitions and the inability of 1-D models to incorporate spatial heterogeneity may be especially problematic.
Recently, we developed 2-D Markov chain geostatistical models to overcome some of these limitations of 1-D Markov models. Preliminary simulation algorithms for modeling 2-D land cover based on the 2-D Markov chain geostatistics have been developed and these algorithms demonstrated significant advantages over traditional methods, such as sequential indicator simulation, in generating more accurate results.
In this research we will further extend the 2-D Markov chain geostatistics to spatiotemporal models of land use change.
2.Modeling distributions of invasive plant species
Land use history and land use change predictions are important components of understanding and forecasting invasive species distributions. Previous studies suggest that land use changes provide opportunities for particular plants to invade an area, that the types of changes that promote one species might inhibit others, and that land use changes contribute not only to invasive species establishment, but also spread. The generation of forest edges and the fragmentation of contiguous forest habitat are particularly important to woody invasive plant richness. Our research will extend previous studies by incorporating prospective land use changes and changes in landscape geometry into species distribution (SDMs) and abundance modeling frameworks.
Invasive species modeling presents a set of challenges, including non-equilibrium with their new environment and multi-scale drivers of species’ distributions and abundance. These challenges have been met with a variety of statistical and other modeling approaches, each of which offers strengths and are complementary. Hierarchical Bayesian (HB) regression models can readily incorporate multi-scale distribution drivers (e.g., local habitat, landscape configuration, climate) into a single model along with spatial dependence in sampling locations and dispersal limitation with Conditional Auto-Regressive or Predictive Point-Process models. Native range information can be incorporated through the use of priors to account for incomplete sampling of potentially suitable conditions in the invasive range. These HB methods naturally include multiple sources of uncertainty, such as uncertainty in the data, the model, and predictions and have outperformed other regression approaches that utilize presence/absence data. In many cases, data are limited to presence-only species occurrences; we have made recent progress modeling presence-only data in a HB regression framework and developed ecologically and statistically informed guidelines for using MaxEnt with many covariates, non-equilibrium conditions, and native range data. Finally, demographically-informed models provide an opportunity to model range dynamics over time. Demographically-informed models demand data that is more difficult and time-consuming to collect than HB regression and MaxEnt models, but can potentially provide greater insight into species abundance patterns. We are actively working to advance this modeling approach by incorporating output from integral projection demographic models (IPMs) into a distribution modeling framework.
Future forecasts of invasive plants in the Northeast have focused on climate change, but land use/land cover change projections and interactions with climate have not yet been explored. In this project, we will investigate forecasts of invasive plants in LISW under land use change predictions using a variety of modeling methods, including HB regression, MaxEnt, mechanistic CA, and IPM-based distribution and abundance models. We will compare and contrast model output among model types to provide analysis of model congruency or disagreement.
Connect With Us